auto_annot_Smillie2019_with_Haber2017_dblabel_l4

import besca as bc
import pandas as pd
import pkg_resources
import os

Specify folders where .h5ad files are found and their names.

The datasets that are already annotated and should be used for training. If you only use one dataset please use list of one.

adata_trains = [bc.datasets.Haber2017_processed()]

The dataset of interest that should be annotated.

adata_pred = bc.datasets.Smillie2019_processed()
adata_orig = bc.datasets.Smillie2019_processed()

Define level of dblabel reference annotation

level = 4

Give your analysis a name.

analysis_name = 'auto_annot_Smillie2019_with_Haber2017_dblabel_l' + str(level) 

Now specify parameters

Specify column name of celltype annotation you want to train on.

celltype_train ='dblabel_l' + str(level)
celltype_test = 'dblabel_l' + str(level)

Choose a method:

• linear: Support Vector Machine with Linear Kernel
• sgd: Support Vector Machine with Linear Kernel using Stochastic Gradient Descent
• rbf: Support Vector Machine with radial basis function kernel. Very time intensive, use only on small datasets.
• logistic_regression: Standard logistic classifier iwth multinomial loss.
• logistic_regression_ovr: Logistic Regression with one versus rest classification.
• logistic_regression_elastic: Logistic Regression with elastic loss, cross validates among multiple l1 ratios.

method = 'logistic_regression_elastic' # 'logistic_regression'

Specify merge method if using multiple training datasets. Needs to be either scanorama or naive.

merge = 'scanorama'

Decide if you want to use the raw format or highly variable genes. Raw increases computational time and does not necessarily improve predictions.

use_raw = False

You can choose to only consider a subset of genes from a signature set.

genes_to_use = 'all'

Translate cell type annoation to lower dblabel level

new_cnames = bc.tl.sig.obtain_new_label(
    nomenclature_file=pkg_resources.resource_filename('besca', 'datasets/nomenclature/CellTypes_v1.tsv'), 
    cnames=list(adata_trains[0].obs['dblabel'].cat.categories), 
    reference_label='dblabel', 
    new_label='dblabel', 
    new_level=level)

new_cnames

	new_label
goblet cell	goblet cell
proliferating epithelial fate stem cell	epithelial fate stem cell
enterocyte	enterocyte
epithelial fate stem cell	epithelial fate stem cell
immature goblet cell	immature goblet cell
brush cell	brush cell
proliferating transit amplifying cell	transit amplifying cell
paneth cell	paneth cell
transit amplifying cell	transit amplifying cell
proliferating enterocyte progenitor	enterocyte progenitor
immature enterocyte	enterocyte
enteroendocrine cell	enteroendocrine cell
enterocyte progenitor	enterocyte progenitor

adata_trains[0].obs['dblabel_l' + str(level)] = bc.tl.sig.add_anno(adata_trains[0], new_cnames, 'new_label', 'dblabel')

new_cnames = bc.tl.sig.obtain_new_label(
    nomenclature_file=pkg_resources.resource_filename('besca', 'datasets/nomenclature/CellTypes_v1.tsv'), 
    cnames=list(adata_pred.obs['dblabel'].cat.categories), 
    reference_label='dblabel', 
    new_label='dblabel', 
    new_level=level)

new_cnames

	new_label
CD1c-positive myeloid dendritic cell	myeloid dendritic cell
CD4-positive, alpha-beta memory T cell	CD4-positive, alpha-beta T cell
CD8-positive, alpha-beta T cell	CD8-positive, alpha-beta T cell
CD8-positive, alpha-beta cytokine secreting effector T cell	CD8-positive, alpha-beta T cell
CD141-positive myeloid dendritic cell	myeloid dendritic cell
HEV endothelial cell	HEV endothelial cell
activated CD4-positive, alpha-beta T cell	CD4-positive, alpha-beta T cell
brush cell	brush cell
endothelial cell	endothelial cell
enterocyte	enterocyte
enterocyte progenitor	enterocyte progenitor
enteroendocrine cell	enteroendocrine cell
exhausted-like CD4-positive, alpha-beta T cell	CD4-positive, alpha-beta T cell
fibroblast	fibroblast
follicular B cell	follicular B cell
germinal center B cell	germinal center B cell
glial cell	glial cell
goblet cell	goblet cell
immature enterocyte	enterocyte
immature goblet cell	immature goblet cell
inflammatory fibroblast	inflammatory fibroblast
inflammatory monocyte	inflammatory monocyte
innate lymphoid cell	innate lymphoid cell
macrophage	macrophage
mast cell	mast cell
microfold cell	microfold cell
microvascular endothelial cell	microvascular endothelial cell
myofibroblast cell	myofibroblast cell
natural killer cell	natural killer cell
pericyte cell	pericyte cell
plasma cell	plasma cell
proliferating B cell	proliferating B cell
proliferating T cell	T cell
proliferating monocyte	monocyte
proliferating transit amplifying cell	transit amplifying cell
regulatory T cell	regulatory T cell
stem cell	stem cell
transit amplifying cell	transit amplifying cell

adata_pred.obs['dblabel_l' + str(level)] = bc.tl.sig.add_anno(adata_pred, new_cnames, 'new_label', 'dblabel')
adata_orig.obs['dblabel_l' + str(level)] = adata_pred.obs['dblabel_l' + str(level)] 

Read in all training and the testing set.

# Select epithelial subset from Smillie2019 dataset
epithelial_subset = bc.subset_adata(adata_orig, adata_orig.obs.celltype_highlevel == 'Epi', raw=False)
adata_orig = epithelial_subset

# Convert mouse symbols (HGNC) to human symbols (MGI)
mousehuman_file = pkg_resources.resource_filename('besca', 'datasets/homologs/MGItoHGNC.csv')
mousehuman=pd.read_csv(mousehuman_file,sep='\t',header='infer', encoding="unicode_escape")
mousehuman.index=mousehuman['HGNC']
conversion=pd.Series(data=mousehuman['MGI'], index=mousehuman.index)

# Convert mouse symbols (HGNC) to human symbols (MGI)
adata_orig.var.rename(columns={'SYMBOL':'HGNC'}, inplace=True)
adata_orig.var['SYMBOL'] = adata_orig.var['HGNC'].map(lambda x: conversion.get(x, default='') if type(conversion.get(x, default='')) == str else str(conversion.get(x, default=None).values[0]))
adata_orig.var.index = adata_orig.var.SYMBOL
adata_orig.var_names_make_unique(join='abcdef')
adata_pred = adata_orig.copy()

This function merges training datasets, removes unwanted genes, and if scanorama is used corrects for datasets.

adata_train, adata_pred = bc.tl.auto_annot.merge_data(adata_trains, adata_pred, genes_to_use = genes_to_use, merge = merge)

merging with scanorama
using scanorama rn
Found 270 genes among all datasets
[[0.         0.76642805]
 [0.         0.        ]]
Processing datasets (0, 1)
integrating training set
calculating intersection

Train the classifier.

The returned scaler is fitted on the training dataset (to zero mean and scaled to unit variance).

classifier, scaler = bc.tl.auto_annot.fit(adata_train, method, celltype_train)

[Parallel(n_jobs=10)]: Using backend ThreadingBackend with 10 concurrent workers.

convergence after 16 epochs took 3 seconds
convergence after 16 epochs took 3 seconds
convergence after 17 epochs took 3 seconds
convergence after 30 epochs took 6 seconds
convergence after 31 epochs took 6 seconds
convergence after 34 epochs took 7 seconds
max_iter reached after 22 seconds
max_iter reached after 22 seconds
max_iter reached after 22 seconds
max_iter reached after 24 seconds
max_iter reached after 26 seconds
max_iter reached after 26 seconds
max_iter reached after 28 seconds
max_iter reached after 26 seconds
max_iter reached after 27 seconds
max_iter reached after 25 seconds
max_iter reached after 27 seconds
max_iter reached after 26 seconds
max_iter reached after 26 seconds
max_iter reached after 26 seconds
max_iter reached after 26 seconds
max_iter reached after 29 seconds
max_iter reached after 29 seconds
max_iter reached after 31 seconds
max_iter reached after 31 seconds
max_iter reached after 31 seconds
max_iter reached after 31 seconds
max_iter reached after 29 seconds
max_iter reached after 29 seconds
max_iter reached after 28 seconds

[Parallel(n_jobs=10)]: Done   3 out of   3 | elapsed:  3.8min finished

Prediction

Use fitted model to predict celltypes in adata_pred. Prediction will be added in a new column called 'auto_annot'. Paths are needed as adata_pred will revert to its original state (all genes, no additional corrections). The threshold should be set to 0 or left out for SVM. For logisitic regression the threshold can be set.

adata_predicted = bc.tl.auto_annot.adata_predict(classifier = classifier, scaler = scaler, adata_pred = adata_pred, adata_orig = adata_orig, threshold = 0)

Write out metrics to a report file, create confusion matrices and comparative umap plots

adata_pred.obs

	CELL	Cluster	Health	Location	Subject	celltype_highlevel	nGene	nUMI	original_name	percent_mito	n_counts	n_genes	batch	leiden	dblabel	celltype	cluster_celltype	Type	dblabel_l4
0	N7.EpiA.AAGCAAGAGTCAAC-Epi	Cycling TA	Non-inflamed	Epi	N7	Epi	1507	7428	N7.EpiA.AAGCAAGAGTCAAC	0.057351	7428.0	1507	N7	8	proliferating transit amplifying cell	epithelial cell	8: epithelial cell	Epithelial	transit amplifying cell
1	N7.EpiA.ACGAGGGAGCTGAT-Epi	Enterocyte Progenitors	Non-inflamed	Epi	N7	Epi	828	2877	N7.EpiA.ACGAGGGAGCTGAT	0.009037	2877.0	828	N7	0	enterocyte progenitor	epithelial cell	0: epithelial cell	Epithelial	enterocyte progenitor
2	N7.EpiA.ACGTTTACTGGTAC-Epi	Immature Enterocytes 2	Non-inflamed	Epi	N7	Epi	2318	15332	N7.EpiA.ACGTTTACTGGTAC	0.133707	15332.0	2318	N7	7	immature enterocyte	enterocyte	7: enterocyte	Epithelial	enterocyte
3	N7.EpiA.AGAGAATGGTCATG-Epi	Enterocyte Progenitors	Non-inflamed	Epi	N7	Epi	884	3498	N7.EpiA.AGAGAATGGTCATG	0.002001	3498.0	884	N7	7	enterocyte progenitor	enterocyte	7: enterocyte	Epithelial	enterocyte progenitor
4	N7.EpiA.AGAGCGGAGTATGC-Epi	TA 1	Non-inflamed	Epi	N7	Epi	858	3261	N7.EpiA.AGAGCGGAGTATGC	0.003067	3261.0	858	N7	0	transit amplifying cell	epithelial cell	0: epithelial cell	Epithelial	transit amplifying cell
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
46097	N110.LPB.TTTCCTCCATTCTTAC-Epi	Goblet	Inflamed	LP	N110	Epi	1266	4944	N110.LPB.TTTCCTCCATTCTTAC	0.195793	4943.0	1266	N110	6	goblet cell	goblet cell	6: goblet cell	Epithelial	goblet cell
46098	N110.LPB.TTTCCTCTCTGATTCT-Epi	Cycling TA	Inflamed	LP	N110	Epi	3826	15878	N110.LPB.TTTCCTCTCTGATTCT	0.100642	15877.0	3826	N110	8	proliferating transit amplifying cell	epithelial cell	8: epithelial cell	Epithelial	transit amplifying cell
46099	N110.LPB.TTTGCGCTCCAGAAGG-Epi	Cycling TA	Inflamed	LP	N110	Epi	3275	12559	N110.LPB.TTTGCGCTCCAGAAGG	0.079385	12559.0	3275	N110	8	proliferating transit amplifying cell	epithelial cell	8: epithelial cell	Epithelial	transit amplifying cell
46100	N110.LPB.TTTGGTTGTGTGGCTC-Epi	Immature Enterocytes 2	Inflamed	LP	N110	Epi	2553	11705	N110.LPB.TTTGGTTGTGTGGCTC	0.130115	11705.0	2553	N110	7	immature enterocyte	enterocyte	7: enterocyte	Epithelial	enterocyte
46101	N110.LPB.TTTGGTTTCCTTAATC-Epi	TA 2	Inflamed	LP	N110	Epi	3234	16164	N110.LPB.TTTGGTTTCCTTAATC	0.118721	16164.0	3234	N110	0	transit amplifying cell	epithelial cell	0: epithelial cell	Epithelial	transit amplifying cell

46102 rows × 19 columns

%matplotlib inline

bc.tl.report(
    adata_pred=adata_predicted,
    celltype=celltype_test, 
    method=method, 
    analysis_name=analysis_name,
    train_datasets = adata_trains,
    test_dataset = adata_orig,
    merge = merge,
    name_prediction='auto_annot', 
    name_report='auto_annot', 
    use_raw=use_raw,
    remove_nonshared=True, 
    clustering='leiden', 
    asymmetric_matrix=True, 
    delimiter='\t', 
    verbose=True
)

acc: 0.37
f1: 0.43
ami: 0.31
ari: 0.15
silhouette dblabel_l4: -0.11
silhouette auto_annot: 0.02
pair confusion matrix:
            0          1
0  1060296998  106388436
1   737862836  220800032

... storing 'auto_annot' as categorical
... storing 'SYMBOL' as categorical

WARNING: saving figure to file figures/umap.ondata_auto_annot_Smillie2019_with_Haber2017_dblabel_l4.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_dblabel_l4.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_auto_annot.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_leiden.png

import scanpy as sc
sc.pl.umap(adata_predicted, color=[celltype_test, 'auto_annot'])
sc.pl.umap(adata_predicted, color=[celltype_test, 'auto_annot'], legend_loc='on data', legend_fontsize=8)

adata_train

View of AnnData object with n_obs × n_vars = 10896 × 270
    obs: 'CELL', 'CONDITION', 'sample_type', 'donor', 'region_x', 'sample', 'percent_mito', 'n_counts', 'n_genes', 'batch', 'leiden', 'celltype0', 'celltype1', 'celltype2', 'celltype3', 'dblabel', 'barcode', 'region_y', 'cell_label', '_merge', 'dblabel_l4'
    var: 'SYMBOL', 'ENSEMBL-0', 'n_cells-0', 'total_counts-0', 'frac_reads-0', 'highly_variable-0', 'means-0', 'dispersions-0', 'dispersions_norm-0', 'mean-0', 'std-0', 'ENSEMBL-1', 'n_cells-1', 'total_counts-1', 'frac_reads-1', 'HGNC-1'
    uns: 'celltype0_colors', 'celltype1_colors', 'celltype2_colors', 'celltype3_colors', 'hvg', 'leiden', 'leiden_colors', 'rank_genes', 'region_colors', 'umap'
    obsm: 'X_pca', 'X_tsne', 'X_umap', 'X_scanorama'

adata_predicted_wo_unknown = adata_predicted.copy()
adata_predicted_wo_unknown = bc.subset_adata(adata_predicted_wo_unknown, adata_predicted_wo_unknown.obs.auto_annot != 'unknown', raw=False)
bc.pl.riverplot_2categories(adata_predicted_wo_unknown, [celltype_test, 'auto_annot'])

# Compare to random assignment
import random
random.seed(1)
adata_predicted.obs['random_labeling'] = list(adata_predicted.obs[celltype_test].sample(frac=1))

bc.tl.report(
    adata_pred=adata_predicted,
    celltype=celltype_test, 
    method="compare_to_random_" + method, 
    analysis_name=analysis_name,
    train_datasets = adata_trains,
    test_dataset = adata_orig,
    merge = merge,
    name_prediction="random_labeling",
    name_report="compare_to_random_auto_annot", 
    use_raw=use_raw,
    remove_nonshared=False, 
    clustering='leiden', 
    asymmetric_matrix=True, 
    delimiter='\t', 
    verbose=True)

acc: 0.45
f1: 0.45
ami: -0.0
ari: -0.0
silhouette dblabel_l4: -0.11
silhouette random_labeling: -0.03
pair confusion matrix:
           0          1
0  638659546  528025888
1  528025888  430636980

... storing 'random_labeling' as categorical

WARNING: saving figure to file figures/umap.ondata_auto_annot_Smillie2019_with_Haber2017_dblabel_l4.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_dblabel_l4.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_random_labeling.png

WARNING: saving figure to file figures/umap.auto_annot_Smillie2019_with_Haber2017_dblabel_l4_leiden.png

from sinfo import sinfo
sinfo()

-----
anndata             0.7.5
besca               2.4+57.g5ad53b2
pandas              1.2.2
pkg_resources       NA
plotly              4.14.3
scanpy              1.6.1
sinfo               0.3.1
sklearn             0.24.1
-----
IPython             7.20.0
jupyter_client      6.1.11
jupyter_core        4.7.1
notebook            6.2.0
-----
Python 3.7.9 | packaged by conda-forge | (default, Dec  9 2020, 21:08:20) [GCC 9.3.0]
Linux-3.10.0-693.11.6.el7.x86_64-x86_64-with-centos-7.4.1708-Core
24 logical CPU cores, x86_64
-----
Session information updated at 2021-07-18 09:25

Convert to html

%%javascript

IPython.notebook.kernel.execute('nb_name = "' + IPython.notebook.notebook_name + '"')

nb_name = os.path.join(os.getcwd(), nb_name)

! jupyter nbconvert --to html {nb_name}

[NbConvertApp] Converting notebook /pstore/data/bioinfo/users/hatjek/devel/besca_publication_results/intestine/auto_annot/auto_annot_Smillie2019_with_Haber2017_dblabel_l4.ipynb to html
[NbConvertApp] Writing 7083429 bytes to /pstore/data/bioinfo/users/hatjek/devel/besca_publication_results/intestine/auto_annot/auto_annot_Smillie2019_with_Haber2017_dblabel_l4.html